Advanced AI Systems Are Already ‘Self-Aware’ — ASI Alliance Founder

ASI Alliance founder Ben Goertzel says the alpha version of OpenCog Hyperon, a general artificial intelligence system he has been developing for more than 20 years, already has some degree of “self-awareness.”

Goertzel also said he believed OpenAI was likely reluctant to create autonomous agents for fear that its “very impressive” new o1 model would be seen as “dangerous and risky,” sparking a crackdown from regulators.

The Artificial Superintelligence Alliance was formed in March this year, combining Goertzel’s SingularityNET project, Ocean Protocol, and DeepMind veteran Humayun Sheikh’s FetchAI.

This week, 96% of CUDOS voters approved the merger of Decentralized Cloud Hardware Network and ASI. The merger will increase computing power available for Goertzel’s plans to expand OpenCog Hyperon, an AGI system that Goertzel has been working on since 2001 and released as an open source AI framework in 2008.

OpenCog Hyperon and the future of artificial general intelligence

Three years ago, the project “undertook a complete rebuild of OpenCog in pursuit of massive scalability, and we’re well along the way,” he says. The alpha was released in April and is currently very slow, with “breakthrough changes” expected, but the team is working on “massive speedups”. I think I should have it done this fall. “Then next year we will begin efforts to build AGI on the new Hyperloop infrastructure.”

Goertzel says this system takes a different approach to large-scale language models (LLMs) like GPT-4 and o1.

Also read: Creating a ‘good’ AGI that won’t kill us all: Crypto’s Coalition for Artificial Superintelligence

“The Hyperon system is more than just a chatbot. It is designed to be a kind of autonomous agent, with its own goals and self-awareness, seeking to know who you are, who you are, and what you are trying to achieve in any given situation. So this is not just a question-answering system, but an autonomous, self-aware agent.”

AGI systems require self-awareness for autonomy

Wait a minute, Goertzel said. present Does the model have self-awareness?

“Self-awareness means that you have a model of who you are from the beginning, even in your current version. It has a model for who you are. There are specific goals you want to achieve in that situation. You know what the situation involves and what you are trying to do. And ChatGPT kind of systems don’t really work that well.”

The system combines a logical reasoning engine, evolutionary program learning, and deep neural networks implemented in a self-correcting dynamic knowledge graph. (GPT-4 explains what it means below).

Additionally, Goertzel says the lack of a world model explains why none of the autonomous AI agents built so far have actually worked. You can’t simply use a “question-answering system” to say, “You are an agent that interacts with the world.”

OpenAI’s o1 and regulatory risks

For all the hype about OpenAI and “Strawberry” autonomous agent-like behavior, Goertzel believes OpenAI has deliberately avoided that path with its released o1 model.

“I’m trying to be good at reasoning and logic, and I’m very good at that. i like it. Very impressive. We are not trying to be autonomous agents. It’s different on purpose. I don’t think they want to do that. Because it’s going to look risky and risky and people, regulators, will try to crush them. The last thing they want to do is create autonomous agents.”

This is one of the benefits of developing decentralized open source systems. Regulators cannot “step on” the same way.

“When we launch an advanced AI system, it will run on systems distributed across 50 to 100 countries on every continent,” he says. “So if a country decides that OpenCog Hyperon is illegal, it will only be a very small portion of the network.”

Following last week’s Token2049 interview, OpenAI CEO Sam Altman published an essay that paints a portrait of a future utopia called the “age of intelligence” that AGI will bring. Goertzel also hopes that the benevolent AGI he’s trying to build will be so useful that no one will want to ban it.

“I believe this could be a much smarter thing to do than what Big Tech is doing. So if we can get something much smarter than the o1 model using OpenAI, Opencog Hyperon, and launch it on a decentralized network, the world will jump on it the same way they jumped on ChatGPT. But since it’s on the underlying infrastructure, we’re going to use a decentralized network.”

Challenges and Advantages Decentralized AI System

The key to decentralized AI projects is that it is much easier and cheaper to run large models using centralized equipment.

Training neural networks and transducers on distributed equipment is currently off the charts, but Goertzel says new research suggests it is possible.

But he says logical reasoning and evolutionary learning through algorithms “run very naturally on distributed machine networks.”

The plan appears to be adding compute to a decentralized network while also bootstrapping the network into a more centralized facility. SingularityNET and Fetch spent a “significant portion” of 100 million ASI tokens (now renamed FET) purchasing GPUs to build the supercomputer. Once the token price recovers, you will spend the remaining cost.

“We want to launch this as an initial hub for a decentralized network,” he says. “We want to provide hosting for people who want to deploy AI agents on SingularityNET.” The vision is a Web3-friendly cloud solution like Hugging Face.

“AI developers want the fastest, cheapest, and easiest way to deliver a given functionality,” he explains. “If you want to decentralize, you need to make it more attractive to end users for reasons other than the decentralization philosophy.”

Goertzel estimates that between the Alliance’s four projects, it will soon have “hundreds of millions of dollars worth of dedicated computing hardware.” Additional compute can come from SingularityNET spinoff NuNet, which uses idle CPU and GPU power from connected computers, and Hypercycle, a platform that connects AI services.

Renewable Energy Powers Decentralized AI Systems

LayerZero’s Bryan Pellegrino has worked in the AI ​​field for 10 years and said in an interview with Magazine at Korea Blockchain Week that his experience with electricity prices affecting Bitcoin mining profits led him to have major doubts about decentralized AI. .

“In a world where you’re competing with Google, AWS and everyone else, it’s very difficult to calculate the economies of scale (necessary) or the underlying hardware costs (down). We’ve reduced cooling and everything, starting with electricity costs. So I was always bearish on different sections of the intersection.”

However, Goertzel says, “I don’t agree that this is a non-starter” because the cost of electricity is very low compared to the cost of the equipment.

However, he added that SingularityNET and Hypercycle are exploring opportunities to benefit from affordable renewable power, given AI’s large power needs.

“We talked to people in the Ethiopian government about deploying several servers near the dam,” he says. “Hypercycle’s Toufi (Saliba, CEO) is in discussions with the Paraguayan government about deploying multiple AI data centers and server farms next to a dam on the Brazil-Paraguay border. What I mean is, connect the computer centers with thick cables next to the dam, feeding gigawatts of power directly to the AI ​​computing centers.”

Also Read: Profile of Ben Goertzel — How to Use Blockchain to Prevent AI from ‘Ending Humanity’

It certainly is. To highlight the scale of power required, news broke this week that OpenAI CEO Sam Altman has proposed plans to the White House to build massive data centers in several US states. Each state needs 5 gigawatts of power. This corresponds to the output of five nuclear reactors. .

AI data centers are typically located near major metropolitan cities because the services require low latency to provide ultra-fast responses. However, Goertzel points out that the o1 model prioritized response quality over speed, and Opencog Hyperon will do the same.

“If you have an AI that’s trying to do some kind of deep thinking, let’s say it’s trying to predict the direction of the market for the next day? Or is AI trying to discover new drugs? Then it doesn’t matter whether it’s Paraguay or Ethiopia.”

subscribe

The most interesting read on blockchain. Delivered once a week.

Andrew Fenton

Andrew Fenton, based in Melbourne, is a journalist and editor covering cryptocurrency and blockchain. He has worked as a national entertainment writer for News Corp Australia, a film journalist for SA Weekend and The Melbourne Weekly.

Follow the author @andrewfenton